OGSA Telecon Minutes -- 15 March 2006

Attendees
---------
	Mark Morgan (minutes)
	Fred Maciel
	Hiro Kishimoto
	Jay Unger
	Ellen Stokes
	Chris Smith
	Mike Behrens
	Stephen Newhouse
	Dejan Milojicic
	Jun Tatemura (CDDLM-WG)
	Heather Kreger

* Minutes from Monday
	- No questions/comments.
	- Approved

* Agenda Bashing
	- F2F Meeting update
	- Joint CDDLM Discussion
	- Timeboxed discussion of the whitepaper from IBM/Intel/etc... (15 min)

* F2F
	- Switch CDDLM and EMS (so that CDDLM is 8am to noon on Thursday)?
		-- Yes, done.

** INTERRUPT **
	- Heather to call in and give over view of IBM/Intel/etc. whitepaper
	  for 10 minutes.
	- Hiro to send invite to Heather for Monday telecon (early in Monday's
	  discussion).
	- Everyone to read whitepaper and come back to discuss next Monday

* Continuation of F2F
	- Mark to send out email about the hotel

* CDDLM
	- Start from Stephen's document or from Hiro's slides.
	- Not sure where Hiro's slides are coming from
	- We have a scenario doc from athens -- first step would be to get
	  CDL guys to give an example document that would be used to deploy
	  the BLAST app. from a repository somewhere -- also a JSDL doc.
	  example.
	- Chris to send example of BLAST workflow (for example DB formatting)
	  to CDL
	- JSDL example for BLAST is for an LSF system.  Not intended to show
	  a provisioning system.  Would probably use the Application element
	  in the JSDL to describe more fully the specifics.
	- Chris suggesting that the current example of JSDL is designed as a
	  very specific example.  Normally you'd have a much more abstract
	  approach with many fields, many options being unspecified and
	  leaving it to other tools to fill these in (after possibly
	  installing things).
		-- It's a typical example right now for a batch system that
		   doesn't do anything for you.
	- DRMAA had the concept of a "job category" and a lot of system's seem
	  to go that way.
	- The general idea is that a client will ask for an application in 
	  the abstract and then CDL will be asked to provide the binary for
	  it.
	- We spent a lot of time last week talking about these slides and we
	  seem to be doing this again.
	- We need consensus -- de we go on with this and evolve this
	  presentation or do we start with Stephen's document?
		-- We expect to have short CDL document that describes how to
		   start up binaries and how to download databases, etc.
		-- CDL to provide this document for F2F
	- Scenario
		-- Client comes to data center and wants to run 10000 searches
		-- He asks a job manager to run 10000 searches
		-- job manager decides to allocate 10 servers for this.
		-- It asks CDL to set up 10 servers.
		-- This takes time, but afterwords the 10000 searches run on
		   it.
	- Dejan to include filesystem mount in the CDL example
		-- Also try to offer more support to the example
		-- When needed?
			--- Atleast the F2F meeting, beforehand better.
			--- Try to provide by EMS call on Monday a week and
			    a half hence.
	- Regarding Environment Variables
		-- Is Environment variable set up part of deployment, or
		   execution.
	- Modify current example to combine the path and tmpdir stuff and put
	  environment into deployment (CDL document).
	- If you want to specify an env. var. into the application execution,
	  then shouldn't that be part of the JSDL?
		-- We should make sure that the distinction between
		   deployment env. variables, and execution env. vars is
		   made clear.
		-- It would be really good to get some of the general
		   use cases and scenarios down.
	- Let's choose one, generic case scenario first and get it down in
	  black and white.
	- Is Env. Var an input to or an output from CDL?
		-- output.
	- Jun to take leadership of next CDDLM meeting.
	- We have a plan.
	- Neither Dejan nor Jun is available on the 27th or the 29th
		-- Actually, Dejan can join on either for the second half, but
		   not the first half.
		-- There isn't a single timeslot that both can make.
	- Put CDL discussion in later part of the 29th telelcon.